LAION drops massive open video dataset with 10 million hours of footage for AI research
80 million videos, 10 million hours: LAION's open dataset beats InternVid by 2.1 points and could shift who can train frontier video models.
Why it matters
LAION's Big Video Dataset removes a major bottleneck for open-weight video model training. With 55 million auto-described clips and legal backing from a 2024 Hamburg ruling on non-commercial research, smaller labs and independent teams now have the raw material to compete with proprietary video models—changing who can build at the frontier.
The key facts
6 to know80 million videos in LAION Big Video Dataset (BVD)
10 million hours of total runtime
55 million auto-described clips
Models trained on BVD beat InternVid benchmark by up to 2.1 percentage points
2024 Hamburg court ruling allows copyrighted content collection for non-commercial research
Positioning LAION BVD as alternative to proprietary video training datasets
Go to the source
The Decoderthe-decoder.com
Publisher excerpt: LAION's Big Video Dataset (BVD) is one of the largest open video datasets for AI research, with 80 million videos, 10 million hours of runtime, and 55 million auto-described clips. Models trained on BVD beat the previous benchmark, InternVid, by up to 2.1 percentage points. Legally, LAION can…